Author: bugman Date: Mon Nov 25 15:39:43 2013 New Revision: 21646 URL: http://svn.gna.org/viewcvs/relax?rev=21646&view=rev Log: Added a MC simulation elimination section to the dispersion chapter of the manual. Modified: branches/relax_disp/docs/latex/dispersion.tex Modified: branches/relax_disp/docs/latex/dispersion.tex URL: http://svn.gna.org/viewcvs/relax/branches/relax_disp/docs/latex/dispersion.tex?rev=21646&r1=21645&r2=21646&view=diff ============================================================================== --- branches/relax_disp/docs/latex/dispersion.tex (original) +++ branches/relax_disp/docs/latex/dispersion.tex Mon Nov 25 15:39:43 2013 @@ -1250,6 +1250,20 @@ When not using the auto-analysis, custom model elimination rules can be defined and used with the \uf{eliminate} user function. +% MC elimination. +%~~~~~~~~~~~~~~~~ + +\subsection{Monte Carlo simulation elimination} + +Just as models can fail, often Monte Carlo simulations will also experience optimisation failures (see Figure~4 of \citet{dAuvergneGooley06} for such a failure in the model-free optimisation space). +The minimum can be warped so much by the data randomisation that a new minimum appears at an unreasonable position in the optimisation space. +Even when the original model optimisation is successful, this can affect a small portion of the simulations. +These must be removed prior to calculating the parameter errors otherwise the errors will be significantly over estimated. +The simulation model failures are outliers which skew the error estimate, introducing a bias. +This can result in parameter error estimates which are too large. +The solution is the use of the \uf{eliminate} user function when Monte Carlo simulations are turned on -- this will automatically deselect simulations rather than spins using the rules from the previous section. +Note that relax is the only software which provides this feature. + % MPI. %~~~~~